Chinese AI company SenseTime has launched SenseNova 5.5, an AI model surpassing GPT-4o across key metrics. Concurrently, Apple, YouTube, KLING, Neuralink, and Google DeepMind have announced significant advancements.
At the 2024 World Artificial Intelligence Conference & High-Level Meeting on Global AI Governance (WAIC 2024), SenseTime hosted its AI Forum titled “AI+: Catalyzing Next-Gen Transformations.” Here, they unveiled the enhanced SenseNova 5.5 Large Model. This update introduces SenseNova 5o, China’s inaugural real-time multimodal model, aligning with GPT-4o’s streaming interaction capabilities.
SenseNova 5.5 has also rolled out a more affordable edge-side large model. It reduces annual per-device costs to just RMB 9.90, facilitating broader deployment. Through ongoing enhancements, SenseTime’s full-stack large model product matrix offers novel solutions for generative applications across diverse sectors. The SenseNova Large Model serves over 3,000 government and corporate clients in technology, healthcare, finance, and programming.
Dr. Xu Li, Chairman and CEO of SenseTime, noted, “This year is pivotal for large models as they transition from unimodal to multimodal formats. We’re enhancing interactivity to meet user demands, driving model development through applications and advanced multimodal streaming interactions. This evolution will transform human-AI interactions significantly.”
Upgrade to SenseNova 5.5
SenseNova 5o processes data across audio, text, images, and videos, delivering a new AI interaction model. Users can interact with SenseNova 5o as if conversing with a human, which is suitable for real-time conversation and speech recognition. This model manages multiple tasks and adjusts responses based on context.
Released in April, SenseNova 5.0 matched GPT-4 Turbo’s capabilities. SenseNova 5.5 showed a 30% performance boost over its predecessor by June. It excels in mathematical reasoning, English proficiency, and command response, matching GPT-4o.
SenseNova 5.5 employs a hybrid cloud-edge architecture, optimizing “Cloud-to-Edge” synergy and lowering inference costs. The model trained on over 10TB of high-quality data, including synthetic reasoning chain data, improves its reasoning capabilities.
To ease enterprise access to the robust SenseNova Large Model, SenseTime launched “Project $0 Go.” This initiative offers new enterprise users migrating from the OpenAI platform a free onboarding package, including 50 million tokens and API migration consulting.
“Cloud-to-Edge” Full-stack Enhancements
SenseTime promotes R&D of edge-side large models, supporting deployment on various IoT devices. By reducing the cost to only RMB 9.90 per device annually, SenseTime aims to boost adoption through cost-effectiveness and accessibility. Currently, over 150 customers are engaged in commercial partnerships.
Additionally, SenseTime upgraded the edge-side large model, launching SenseChat Lite-5.5. This model features reduced inference time and a 15% speed increase, enhancing overall performance.
SenseTime also introduced an edge-side model product matrix, including specialized models like the SenseChat Mini Writing Assistant, the Summary Assistant, and the Encyclopedia Assistant. These models offer enhanced performance for specific scenarios, allowing customers to tailor solutions to their needs.
New Additions to SenseNova’s Suite of Applications
As part of SenseNova 5.5, SenseTime released Vimi, a controllable AI avatar video generator. Using just a photo, Vimi creates video clips with precise control over facial expressions and body movements, producing realistic changes in lighting, shadows, and backgrounds. Vimi supports video generation for up to one minute without quality loss, which is ideal for entertainment and interactive uses.
Building on the “Cloud-to-Edge” full-stack model, SenseTime continues to advance generative AI applications under the SenseNova Large Model Series, meeting broader user needs and supporting digital transformation across industries.
The SenseTime Raccoon Series, including the SenseNova AI Native productivity tool, has also been upgraded. The Code Raccoon (Consumer Edition) now responds five times faster and offers 10% more coding precision, with enhanced model capabilities and a richer plugin set. Office Raccoon now features a consumer-facing webpage and a WeChat mini-app, enabling efficient file uploading, analysis, and processing directly within WeChat.
As SenseTime marks its 10th anniversary, it remains committed to expanding the SenseNova industry ecosystem, empowering more businesses and communities in their digital journeys through innovative AI solutions.